Page 1
Exploring Cognitive Distraction of Galvanic Skin
Response while Driving: An Artificial
Intelligence Modeling
Chiang-Yu Cheng School of Big Data Management, Soochow University, Taiwan
Email: [email protected]
Wesley Shu International Business School, Xi’an Jiaotong-Liverpool University, Suzhou, China
Email: [email protected]
Han-Ping Tsen Department of Information and Management, National Central University, Taiwan
Email: [email protected]
Abstract—It is quite often that we hear fatal traffic accidents
due to driver’s distraction. Car manufactures and
researchers are therefore putting their efforts into car safety
protection mechanism. However, the application of car
safety protection mechanism is frequently hindered by its
limitations, such as drivers’ privacy or the high cost of its
deployment and each of which leads to rare applications of
car safety protection specifically in the field of non-
autonomous cognitive distraction. This research proposal
intends to apply the sensor of Galvanic Skin Response (GSR)
to measure drivers’ non-autonomous cognitive distraction
due to the blood glucose variation of diabetes. SVM-RFE
will be adopted as the major algorithm to create an alert
mechanism with the artificial intelligence concept of
supervised machine learning. The researched human-
machine sense interaction mechanism can be able to embed
into the car computer so that it can detect drivers’
physiological changes during diabetes outbreak and then
raise advisable alert and intervention accordingly.
Index Terms—cognitive distraction, galvanic skin response,
artificial intelligence, supervised machine learning
I. INTRODUCTION
There are many companies in China devote in the
development of unmanned vehicle, such as Baidu,
Tencent and Alibaba [1]. These practitioners are
undoubtedly wanting to commercialize their driverless
technology as soon as possible to realize a better driving
experience for humans. However, driverless technology
needs to rely on many sensors to collect massive sensing
data with extremely difficulty processing so that
driverless technology can prevent traffic accidents that
may occur during the driving process [2]. Goodrich &
Schultz (2008) also asserted that “human-machine
sensing interaction” will be the key factor that affects the
Manuscript received August 29, 2019; revised December 24, 2019.
success of unmanned driving technology [3], because
whether or not the vehicle computer can seamlessly
bridge the interaction between the driver and the vehicle
should be vital to driverless technology.
A. Classification of Driving Distraction
According to the survey reported by National Highway
Traffic Safety Administration that nearly 80% of traffic
accidents were caused by driver distractions [4]. Azman
et al. (2010) stated that driving distraction can be divided
into three categories, including visual distraction, manual
distraction and cognitive distraction [5]. Visual
distraction represents the driver moves her/his eyesight
from the road to somewhere (e.g., finding a parking place,
checking a road sign), while manual distraction refers to
the driver’s hands leave away from the steering wheel
and to do something (e.g., adjusting the seat, smoking).
As for cognitive distraction, it means that the driver is
mentally unable to focus on the driving task (e.g.,
communicating with passengers, thinking about
something unrelated to the driving or physical
discomfort).
B. Detection of Driving Distraction
Prior studies have figured out the way in which
human-machine sensing interaction can be applied to
protect driving safety when the driver has distractions.
For example, Rongben et al. (2004) speculated the
driver’s cognitive distraction by using a camera to
capture the driver’s lip data and then applied neural
network to model the driver’s conversation or yawn [6].
Harada et al. (2014) elaborated the driver’s visual
distraction by using a camera to capture images of the
driver’s eye movements and also applied neural networks
as the modeling algorithm [7]. Sathyanarayana et al.
(2008) used a gravity accelerometer with a gyroscope to
collect the driver’s head and leg movements to determine
if the driver has manual distraction [8]. Although these
© 2020 J. Adv. Inf. Technol.
Journal of Advances in Information Technology Vol. 11, No. 1, February 2020
35doi: 10.12720/jait.11.1.35-39
Page 2
distraction detections are valid, they are practically
inapplicable due to the facts that the driver’s privacy can
be violated by the detecting camera and the distraction
detecting equipment can be cost a lot.
C. Lack of Detecting the Cognitive Distraction
It can be found that most of the detections about
driving distraction focus on the type of visual distraction
and manual distraction, however, the cognitive detection
is rarely to be mentioned. Unfortunately, cognitive
detection happened in various situations, including
excessive drinking, fatigue driving and physical
discomfort. When encountering these situations which are
irresistible, the mechanism of early warning is necessary
to prevent the car accidents caused by the cognitive
distraction. Therefore, some researchers devote to the
detection of cognitive distraction. Almahasneh et al.
(2014) apply the Brainwave Scanner to collect ECG data
from drivers, but the sensing equipment is heavy and
expensive [9]. Dehzangi et al. (2018), however, applied
Galvanic Skin Response (GSR) to overcome this research
limitation [10]. GSR is an electrical skin conductivity
sensor which collects data from the interaction between
human psychological state (i.e., distraction) and her/his
surrounding environment (i.e., driving). In addition, GSR
uses patch to measure skin conductance and therefore it
has relatively low deployment cost than other distraction
detection equipment. Most importantly, the patch of GSR
does not violate the subject’s privacy in that it has no
personally identifiable information. Nevertheless, the
testing situation in his research is autonomous (talking on
the phone or sending the message), so the result cannot be
applied in the situations which are irresistible (physical
discomfort or fatigue driving).
D. Difficult to Collect Data during Driving
There were few researchers applying the GSR device
to detect the cognitive distraction, and the possible reason
is that the variation of GSR data from human is really
complicated. Furthermore, once researchers can explain
the variation of GSR data, it can be used to identify the
specific situations. However, it is trackable that the
cognitive distraction is triggered by the physical
discomfort from drivers. For example, a person who has
the headache and loss the visual, it may triggered by the
dropping of blood glucose. Therefore, if the driver with
diabetes whose blood glucose rapidly change without the
mechanism of early warning in the car, it may cause the
serious car accident. Frier (2000) also verify that diabetic
drivers may occur cognitive distraction due to the
variation of the blood glucose [11]. However, with the
limitation that the detection of blood glucose is invasive,
it is difficult to collect data during the driving situation.
This research proposal applies GSR to evaluate the
driver’s cognitive distraction while driving. Specifically,
drivers with diabetes will be selected as the research
subjects to confirm whether or not their GSR values
(physiological state) can be affected by the blood glucose
variance. Evaluating cognitive distraction with diabetes
drivers is quite different and rigorous than that of
Dehzangi et al. (2018) even though their research mainly
focuses on cognitive distraction as well [10]. As a result,
drivers’ irresistible distractions (e.g., distraction caused
by disease attack) should be more harmful and
unpredictable than autonomous distractions (e.g., using
mobile phone). Therefore, the current research proposal
combines the advantages of low cost, privacy protection,
and practical feasibility to attempt to surpass the
measurement obstacle of drivers’ irresistible cognitive
distractions in the past research.
II. LITERATURE REVIEW
A. Signals Detection Method from Human Body
Detecting the signals from human body is the
foundation to understanding the various human reactions.
There are many methods to detect the human signals,
including ECG (Electrocardiography), EMG
(Electromyography) and GSR (Galvanic Skin Response).
ECG is a test that records the electrical activity of
participant’s ticker through small electrode patches that a
technician attaches to the skin of his chest, arms, and legs.
For example, Steinvil et al. (2011) applied ECG to collect
data from athletes in the break time and exercise time
separately, aiming to explore whether the detection of
ECG can reduce the possibility of sudden cardiac death in
athletes [12]. However, due to ECG is a multiple-point
detection, it is not suitable for the driving situations (e.g.
ankles sensing may affect the driving safety). As for the
EMG detection, it is diagnostic procedure to assess the
health of muscles and the nerve cells that control the
motor neurons. For its practical application, Donovan et
al. (2015) applied the EMG to explore the muscle status
from the participant who has Chronic Ankle Instability
during his/her exercise time [13]. However, EMG is not a
suitable method to detect the cognitive distraction during
driving situation because of its invasive detection may
cause uncomfortable feeling for drivers. Last but not least,
GSR detection refers to the changes in sweat gland
activity which are reflective of the intensity of
participants’ emotional state. Horvath (1978) proposed
that human skin is a conductor, so that can judge the
physiological (GSR data) and psychological (feeling
happy or sad) reaction of participant by the variation of
sweat secretion [14]. However, GSR data is variety of
variation, so it is difficult to define how specific
waveform of GSR data can correspond to the specific
psychological reaction.
Due to the low cost and convenient usage [10], we
attempt to apply GSR as the method to collect data from
drivers, besides, we can embed the GSR device into the
steering wheel, so that can reduce the Hawthorne Effect.
After that, we try to define the specific GSR data by
supervised machine learning, so that can correspond to
the specific psychological reaction.
B. The Applications of GSR
Due to the low cost and the convenient usage of the
GSR, researchers applied it as the method to collect data
in many different applications. It can roughly classify the
application fields of GSR into the psychological status
© 2020 J. Adv. Inf. Technol.
Journal of Advances in Information Technology Vol. 11, No. 1, February 2020
36
Page 3
mining and the physiological status confirmation. For the
application of the psychological status mining (explore
the psychological status from the participant), Kurniawan
(2013) detect stress from speakers with the signals of
GSR data, aiming to provide the better stress
management [15]. For the application of the
physiological status confirmation, Majumder (2017)
considered that sensors paly the key role in the smart
home [16]. In order to know the activity of elders at home
and ensure their safety, it can be realized with the
wearable device embedded the GSR sensor.
In this study, we attempt to across the applications
between the psychological status mining and
physiological status confirmation. We collect the GSR
data to detect the psychological status from drivers during
the driving situation, then modeling for the identification
of the cognitive distraction caused by their physical
discomfort according to the GSR data.
III. RESEARCH METHOD
There are five steps in our research method (Fig. 1),
including raw data collection, data classification, data
segmentation and analysis, features extraction, features
selection and modeling.
Figure 1. The research steps
A. Raw Data Collection
1) Participants and conditions
In order to detect the occurrence of the irresistible
cognitive distraction from drivers, we select four diabetic
patients as the participants in this study. The main reason
is that diabetes is a chronic disease and patients’ blood
glucose fluctuation may affect their driving safety [17].
With the trend of the age about the diabetic patients
becomes younger [18] and the average age is 59.9 [19],
we select the participants in male between the age of 30-
40. Besides, we start to collect GSR data from drivers as
a dinner time approached, aiming to capture the process
that blood glucose declines. Finally, in order to identify
the occurrence of the cognitive distraction by the
supervised machine learning, we treat four diabetic
patients as experimental groups, and also select four
healthy people as the control group.
2) Deplyment of GSR decive
In this study, we drive the GSR sensor by Arduino, a
single-chip microcomputer. Arduino is famous for
Makers recently, and users can control their sensors with
its exclusive Integrated Development Environment (IDE).
GSR is a wearable device (Fig. 2 [20]), it is composed by
a digital to analog converter and two finger sleeves with
electrode. Besides, we attempt to embed the GSR device
into the steering wheel to reduce the Hawthorne Effect,
so that can realize the data collection only by holding the
steering wheel for participants.
Figure 2. GSR device [20]
B. Data Classification
The collected data will divide into two types, including
tonic components and phasic components. The former
represents a long-term skin conductance whereas the
latter refers to short-term skin conductance. In this study,
due to the occurrence of the cognitive distraction belongs
to the specific events arising the variation of the skin
conduction, we only select phasic components as the
modeling target afterwards.
C. Data Segmentation and Analysis
With the characteristic that phasic components varies
with time and affected by multiple factors, we adopt
Spectral Analysis to evaluate its features. In order to
explore the time-frequency features from the
experimental group (diabetic patients) and control group
(healthy drivers) based on the phasic components, we
adopt Time-Frequency Analysis to analyze the spectral
and temporal distribution. Dehzangi et al. (2018) also
used the method stated above to analyze the phasic
components in three different situations [10], including
the general driving situation, driving situation with
talking on the phone and driving situation with sending
the message. The outcome is shown in Fig. 3 below.
Besides, in order to enable data responds in the short
time, we adopt the method of sliding windows to segment
data. Each windows include 5 seconds, and we extract the
fifth data by superimposing four data in the front.
© 2020 J. Adv. Inf. Technol.
Journal of Advances in Information Technology Vol. 11, No. 1, February 2020
37
Page 4
Figure 3. Output of the time-frequency analysis
D. Features Distraction
In each window we will extract eight features for
machine learning afterwards, including Average,
Variance, Accumulation, Maximum, Minimum, Peak
number, Average of peak’s aptitude and Auto-regression.
They are listed below as Table I.
TABLE I. LIST OF FEATURES
Feature Name Feature Description
Average the average of the phasic components in a window
Variance the average distance between each sample
points and Average in a window
Accumulation Sum of the phasic components in a window
Maximum The maximum value of phasic components in a
window
Minimum The minimum value of phasic components in a
window
Peak number If both side of value less than the current value, counted as a peak
Average of
peak’s aptitude The average of the peak’s aptitude in a window
Auto-regression An output by an AR model based on the data in a window
E. Features Selection and Modeling
In this study, we will apply SVM-RFE (Support Vector
Machine-Recursive Feature Elimination) to select the
features. First, it will classify the data into two groups
according to the given features with SVM algorithm, then
the features will be sorted by the received score. Next, the
feature with lowest score will be eliminated, and the
remaining features will continue to train the model. After
several iterations, we finally select three target features
for modeling and achieve the purpose of identifying the
cognitive distraction in a high accuracy.
IV. EXPECTED RESULTS
With skin conductance sensing and machine learning,
the current study is expected to be able to detect the
decline of blood glucose in diabetes sufferers during their
driving. This will not only solve the detection problem of
diabetes drivers while driving, but also practically
contribute to an artificial intelligence warning mechanism
that reduces traffic accidents when the drivers come
across to irresistible distraction. More specifically, the
expected results of this study include: (1) confirming the
GSR difference between healthy and diabetes drivers. (2)
Applying GSR to the driving scenario of irresistible
distraction. (3) Creating artificial intelligence machine
learning model in the field of driverless technology. (4)
Specifying cognitive distraction measurement in both
academy and practice.
CONFLICT OF INTEREST
The authors declare no conflict of interest.
AUTHOR CONTRIBUTIONS
All the authors have contributed to this study equally.
REFERENCES
[1] K. Jia, M. Kenney, J. Mattila, and T. Seppala, “The application of artificial intelligence at Chinese digital platform giants: Baidu,
Alibaba and Tencent,” February 2018. [2] R. Benenson, T. Fraichard, and M. Parent, “Achievable safety of
driverless ground vehicles,” in Proc. 10th International
Conference on Control, Automation, Robotics and Vision, 1993, pp. 515-521.
[3] M. A. Goodrich and A. C. Schultz, “Human–robot interaction: A survey,” Foundations and Trends® in Human–Computer
Interaction, vol. 1, no. 3, pp. 203-275, January 2008.
[4] NHTSA, National Highway Traffic Safety Administration. (2017). [Online]. Available: https://www.nhtsa.gov/risky-
driving/distracted-driving [5] A. Azman, Q. Meng, and E. Edirisinghe, “Non-intrusive
physiological measurement for driver cognitive distraction
detection: Eye and mouth movements,” in Proc. 3rd International Conference on Advanced Computer Theory and Engineering,
August 2010, vol. 3, p. 595. [6] W. Rongben, G. Lie, T. Bingliang, and J. Lisheng, “Monitoring
mouth movement for driver fatigue or distraction with one
camera,” in Proc. the 7th International IEEE Conference Intelligent Transportation Systems, October 2004, pp. 314-319.
[7] T. Harada, H. Iwasaki, K. Mori, A. Yoshizawa, and F. Mizoguchi, “Evaluation model of cognitive distraction state based on eye
tracking data using neural networks,” International Journal of
Software Science and Computational Intelligence, vol. 6, no. 1, pp. 1-16, January 2014.
[8] A. Sathyanarayana, S. Nageswaren, H. Ghasemzadeh, R. Jafari, and J. H. Hansen, “Body sensor networks for driver distraction
identification,” in Proc. IEEE International Conference on
Vehicular Electronics and Safety, September 2008, pp. 120-125. [9] H. Almahasneh, W. T. Chooi, N. Kamel, and A. S. Malik, “Deep
in thought while driving: An EEG study on drivers’ cognitive distraction,” Transportation Research Part F: Traffic Psychology
and Behavior, vol. 26, pp. 218-226, August 2014.
[10] O. Dehzangi, V. Rajendra, and M. Taherisadr, “Wearable driver distraction identification on-the-road via continuous
decomposition of galvanic skin responses,” Sensors, vol. 18, no. 2, p. 503, January 2018.
[11] B. M. Frier, “Hypoglycemia and driving performance,” Diabetes
Care, vol. 23, no. 2, pp. 148-148, February 2000. [12] A. Steinvil, T. Chundadze, D. Zeltser, O. Rogowski, A. Halkin, Y.
Galily, H. Perkuk, and S. Viskin, “Mandatory electrocardiographic screening of athletes to reduce their risk for sudden death: Proven
fact or wishful thinking,” Journal of the American College of
Cardiology, vol. 57, no. 11, pp. 1291-1296, March 2011. [13] L. Donovan, J. M. Hart, and J. Hertel, “Effects of 2 ankle
destabilization devices on electromyography measures during functional exercises in individuals with chronic ankle instability,”
Journal of Orthopaedic and Sports Physical Therapy, vol. 45, no.
3, pp. 220-232, March 2015. [14] F. Horvath, “An experimental comparison of the psychological
stress evaluator and the galvanic skin response in detection of deception,” Journal of Applied Psychology, vol. 63, no. 3, pp. 338-
344, June 1978.
[15] H. Kurniawan, A. V. Maslov, and M. Pechenizkiy, “Stress detection from speech and galvanic skin response signals,” in Proc.
the 26th IEEE International Symposium on Computer-Based Medical Systems, June 2013, pp. 209-214.
[16] S. Majumder, E. Aghayi, M. Noferesti, H. Memarzadeh-Tehran, T.
Mondal, Z. Pang, and M. Deen, “Smart homes for elderly healthcare—Recent advances and research challenges,” Sensors,
vol. 17, no. 11, p. 2496, October 2017.
© 2020 J. Adv. Inf. Technol.
Journal of Advances in Information Technology Vol. 11, No. 1, February 2020
38
Page 5
[17] W. Q. Sturner and A. Sullivan, “Hypoglycemia as the responsible factor in a truck driver accident fatality,” Journal of Forensic
Science, vol. 28, no. 4, pp. 1016-1020, October 1983.
[18] L. J. Weaver and S. V. Madhu, “Type 2 diabetes and anxiety symptoms among women in New Delhi, India,” American Journal
of Public Health, vol. 105, no. 11, pp. 2335-2340, November 2015. [19] G. Levterov, B. Levterova, and M. Orbetzova, “Correlates of
healthrelated quality of life in Bulgarian patients with type 2
diabetes mellitus,” in Proc. 19th European Congress of Endocrinology, May 2017, vol. 49.
[20] MySignals GSR Sensor (Galvanic Skin Response). (2016). [Online]. Available:
https://www.youtube.com/watch?v=qCs5w26wABk&feature=you
tu.be
Copyright © 2020 by the authors. This is an open access article distributed under the Creative Commons Attribution License (CC BY-
NC-ND 4.0), which permits use, distribution and reproduction in any
medium, provided that the article is properly cited, the use is non-commercial and no modifications or adaptations are made.
Chiang-Yu Cheng was born in Taoyuan,
Taiwan on November 5, 1979. He graduated
from Department of Information and Management, National Central University,
Taiwan and received his PhD in 2011. He is presently an Assistant Professor in the School
of Big Data Management, Soochow University,
Taiwan. His research interests include Online Traffic Analysis, Big Data Analysis and IoTs
Programming.
Wesley Shu is a PhD graduated from Department of Management Information
System, University of Arizona, USA in 1998.
He is presently a Professor in the Xi’an Jiaotong-Liverpool University, International
Business School, Suzhou China. His research interests include Information Economics,
Fintech, Management Innovation and
Blockchain.
Han-Ping Tsen was born in Taipei, Taiwan on
August 31, 1997. He is a graduate student in Department of Information and Management,
National Central University, Taiwan. His
research interests include Mobile Traffic Analysis, IoTs Deployment and Big Data
Analysis.
© 2020 J. Adv. Inf. Technol.
Journal of Advances in Information Technology Vol. 11, No. 1, February 2020
39